Bubbles, gullibility, and other challenges for economics, psychology, sociology, and information...

8/9/2019 Bubbles, gullibility, and other challenges for economics, psychology, sociology, and information sciences

1/46

Bubbles, gullibility, and other challenges for

economics, psychology, sociology, and informationsciences

Andrew Odlyzko

School of Mathematics

University of Minnesota

Minneapolis, MN 55455, USA

[email protected]

http://www.dtc.umn.edu/odlyzko

Preliminary version, August 5, 2010

Abstract. Gullibility is the principal cause of bubbles. Investors and the gen-eral public get snared by a beautiful illusion and throw caution to the wind.Attempts to identify and control bubbles are complicated by the fact that theauthorities who might naturally be expected to take action have often (espe-cially in recent years) been among the most gullible, and were cheerleadersfor the exuberant behavior. Hence what is needed is an objective measure ofgullibility.

This paper argues that it should be possible to develop such a measure.Examples demonstrate, contrary to the efficient market dogma, that in some

manias, even top-level business and technology leaders do fall prey to collectivehallucinations and become irrational in objective terms. During the Internetbubble, for example large classes of them first became unable to comprehendcompound interest, and then lost even the ability to do simple arithmetic, tothe point of not being able to distinguish 2 from 10. This phenomenon, togetherwith advances in analysis of social networks and related areas, points to possibleways to develop objective and quantitative tools for measuring gullibility andother aspects of human behavior implicated in bubbles. It cannot be expectedto infallibly detect all destructive bubbles, and may trigger false alarms, butit ought to alert observers to periods where collective investment behavior isbecoming irrational.

The proposed gullibility index might help in developing realistic economicmodels. It should also assist in illuminating and guiding decision making.

1 Introduction

Current investigations of the great financial crash of 2008 concentrate on issues that areancillary, such as banker bonuses and where derivative trading should take place. Strangely(but conveniently for many of those involved) these investigations do not delve into thequestion of whether this crash could have been foreseen and prevented, nor into the funda-mental cause of the crash, namely the extensive gullibility that led to the extreme bubble


2/46

2 Andrew Odlyzko

we have experienced. What characterizes bubbles is a rise in valuations of some class ofassets, and it is the collapse of those valuations that leads to subsequent pain. The incen-

tives and institutions involved in bubbles have varied over the centuries, and will surelyvary in the future. But bubbles we have had for ages, in a variety of settings.

Bubbles (interpreted here in the popular term to mean investment manias that lead torising prices of some assets and then to a collapse) are not always easy to identify. However,as was shown by numerous hedge fund managers who earned fortunes, this was possiblein the bubble that crashed in 2008. This will be discussed in more detail in Section 11.Some other bubbles could be identified a priori much more easily. Yet claims continue tobe made by powerful and respected authorities, in particular by Alan Greenspan [21], thatbubbles are impossible to spot. (This is just one example of the gullibility that facilitatesrise of investment manias.)

The gullibility responsible for the real estate and financial industry bubble that col-lapsed in 2008 was widespread and deep. It involved minimum-wage earners buying ex-pensive houses using mortgages they did not understand, as well as bankers issuing suchmortgages. And of course there were all the people who trusted Bernie Madoff with largesums of money, including many financially sophisticated professionals acting as trusteesfor charitable institutions. (For more examples and a discussion, see Section 10.) The gen-eral opinion is that the level of gullibility was much higher during this bubble (as well asduring other bubbles) than normal. Unfortunately this is a subjective judgment, and theproposal of this paper is to develop a quantitative gullibility index, a formal measure ofthis phenomenon.

An objective measure of gullibility is especially important because we face a variant ofthe old Qui custodiet ipsos custodes? question of Juvenal: Who will watch the watch-men? The economic policy makers and regulators in almost all countries over the lastcouple of decades were enthusiastic practitioners of the see no bubble, hear no bub-ble, speak no bubble philosophy. Consider just the three most prominent professionaleconomists among recent powerful economic policy makers in the U.S., Alan Greenspan,Ben Bernanke, and Larry Summers. As is sketched in Section 10, they not only contributedto the financial debacle through their decisions, but in addition they discouraged investiga-tions by others into investment manias. They proclaimed that bubbles cannot be identified.What they then proved conclusively (in the two recently bubbles, the Internet one and themore recent real estate and finance debacle) was that they are unable to identify bubbles,

even giant ones. Further, they still proclaim, in the face of overwhelming evidence, thatthose bubbles could not have been identified. Yet, they (or, to be more precise, two of thethree) are entrusted with guarding our financial system from future disasters of this type.Were Samuel Johnson to come alive, he might be tempted to come up with a phrase moreexpressive than the triumph of hope over experience to describe the situation.

On the other hand, Samuel Johnson was an astute man, and after observing our societyhe might conclude that the selection of the most gullible for top economic policy decisionsis not an accident. There is an apocryphal story about Otto von Bismarck. Somebody oncesupposedly asked him, How can you tolerate that Baron X in charge of the Ministry of Y?He is so stupid! To which the Iron Chancellor replied, But can you find somebody more


3/46


4/46

4 Andrew Odlyzko

During the Internet bubble, the two biggest real investment disasters (i.e., involvingactual outlays by companies, as opposed to changes in stock market valuations) were

the construction of new long-haul fiber optic networks in the U.S. at at a direct cost ofapproximately $100 billion, and the European 3G spectrum auction in which participat-ing wireless service providers paid approximately e100 billion2. The Economist wrote ofthem [58] that [b]oth of these episodes are now regarded as embarrassing collective hallu-cinations over which the industry prefers to draw a veil. This veil, combined the studiousefforts by economists (with Messrs. Bernanke, Greenspan, and Summers in the lead) andother observers not to notice even the existence of the veil, has allowed Alan Greenspan,for example, to claim with a straight face that bubbles cannot be identified before theyburst [21].

Collective hallucinations in investment settings have not attracted much serious schol-

arly attention in recent years. Mackays Extraordinary Popular Delusions and the Madnessof Crowds continues to be popular, but it is based on unreliable secondary sources, andso cannot be taken seriously. On the other hand, there has been an explosion of researchin a variety of fields, such as anthropology, psychology, sociology, political science, neuroe-conomics, and behavioral economics, which demonstrates how frequently human behaviordeparts from the Homo economicus utility-maximizing assumptions of classical economics.In the interests of brevity, they will not be discussed here, but a a good collection of refer-ences to some of this work can be found in [2]. This paper argues that many of the toolsand insights from these areas should be applied to study collective hallucinations. Themain thrust of this paper, though, is on showing, in simple terms, without resorting to anytechnical results or language, that sometimes even the foremost business and technology

leaders are subject to mass delusions to an extent that drastically skews their investmentdecisions3. Moreover, at least occasionally this effect can be quantified in objective terms.

There is considerable anecdotal evidence for the important influence of mass delusionson modern decision making. Managers and financial analysts accused of improper behaviorduring the Internet bubble often tried to argue, in defending themselves during the classaction lawsuits filed afterwards, that normal standards should not apply to decisions madein the frenzied atmosphere at the height of a mania. Those were obviously self-servingclaims, but they are supported by testimony by other, less interested participants in thathistorical episode. Some are cited in Haackes book Frenzy [23]. Even more support forthis view can be obtained from other sources. For example, Neil Barton, the London-based

technology analyst for Merrill Lynch during the 1980s and 90s, retired in 1999. He recalls(private communication) that he was suspicious of the developing market and economicsituation and began to doubt even his own judgment. He was uncomfortable with thevery fast-growing buy-side, where both new and established investment managers reactedirrationally to even mildly negative reports.

Still, even Bartons is just anecdotal evidence, and what appears needed is an objectivemeasure of the degree of irrationality that develops during bubbles. The aim of this paper isto point out ways that such a measure might be developed. This investigation also leads tonew ways of looking at economic activity, and suggests that bubbles may be an inevitablebyproduct of the processes that produce economic growth.


5/46

Bubbles and gullibility 5

The failure of conventional macroeconomics, demonstrated in the recent financial crashof 2008, has stimulated interest in considering other approaches. Lo and Mueller [37] argue

economics has been unrealistic in trying to achieve physics-like standards of predictability.Akerlof and Shiller, in their recent book Animal Spirits [1], argue even more strongly thathuman psychology is the driving force behind economic behavior. They base much of theirarguments on, and take their title from, John Maynard Keynes. They go beyond Keynesby also emphasizing the importance of stories (what I will call tales) that spread throughsociety and inspire economic activity.

The core of this paper involves a whale of a tale, the myth of Internet traffic doublingevery 100 days. It was a key inspiration for the entire Internet bubble. The Madoff Ponzischeme appears to have involved somewhere between $20 and $65 billion. About $20 billionis what the gullible investors put into it, drawn by the beguiling tales of Madoffs magical

trading system. And $65 billion was about the sum of the largely imaginary balancesthat Madoff created. The Internet traffic doubling every 100 days tale led to real directinvestments that easily exceeded $100 billion, and likely was several times that, and tostock market valuations that were in the trillions of dollars. Thus this tale was far moreinfluential than the Madoff one. It was also far more transparently nonsensical than theMadoff Ponzi scheme, and involved important business and technology leaders not onlywillingly suspending their disbelief, but losing the ability to do simple arithmetic.

This paper is based on material from a book-in-progress, tentatively entitled BeautifulIllusions and Credulous Simplicity: Technology Manias from Railroads to the Internet andBeyond, [49], which is devoted to comparing the Internet bubble to the British Railway

Mania of the 1840s and making projections about future technomanias. The Railway Maniawas the largest technology mania in history, when measured in real investments as a fractionof the economy. It pulled in as investors such famous personalities as Charles Darwin,John Stuart Mill, and the Bronte sisters. Several manuscripts from that project, concernedprimarily with early railroads, are available on my home page, in particular [47,48]. Muchof the material about the telecom debacle during the Internet mania is based on previouslypublished papers, in particular [8,46].

The first half of the 19th century may seem remote to us. But it deserves special at-tention, as that was the formative period of modern corporate capitalism, when many ofour most important laws, regulations, and institutions, and our entire institutional culturedeveloped. Interestingly enough, one finds among some observers of that time an under-

standing of animal spirits deeper than that of Keynes. Much of it is even deeper thanthat of Akerlof and Shiller, with their discussion of the importance of tales. It involvedan appreciation of the role of the promoters who concoct, embellish, and propagate suchtales. After any crash, they are often referred to scornfully as Pied Pipers and snake oilsalesmen, but that is a very one-sided view. They included people such as Benjamin Dis-raeli during the British investment bubble of the mid-1820s, see Chapter 5 of [47]. Theircontributions towards stimulating economic activity through creation and propagation ofbeautiful illusions should not be underestimated, just as the importance of technical andmanagerial competence and of honesty should not be overestimated. Further, many of themcannot be placed easily on the fool to rogue scale. Often, as was the case with Disraeli,


6/46

6 Andrew Odlyzko

they believe enough of their tales (drink their own Kool-Aid in a modern phrase) tolose their fortunes in the bubbles they help inflate. In general, the mental states of various

participants at times of mania peaks are indeed different from normal ones (and deservingof more careful studies using modern tools). This was appreciated by some of the moreperceptive of the early Victorian observers. And so was the value of credulous simplicityamong the public and among decision makers.

The next section has a brief discussion of innumeracy, which is intimately connected togullibility, and offers one of the most promising approaches to a quantitative measure ofthis characteristic. Section 3 introduces the telecom bubble and the key Internet trafficdoubling every 100 days tale. There is a briefl discussion of the significance of this tale, andwhat reactions to it signified about the listeners ability to understand compound interest.Then sections 4 through 8 delve in more detail into this whale of a tale and demonstratethat most of the accomplished and supposedly sophisticated professionals involved withthe telecom sector as either investors or employees, or reporters, were acting in a mentalhaze. They were unable to tell the difference between growth rates of 100% (2x) per yearand 1,000% (11x) per year, even when those affected the most important measure of theirindustrys health. It is rather intriguing that the only person at the time who left anypublic record of recognizing the difference existed, and that it mattered. was George Gilder.Exalted as the foremost guru of the Internet revolution while share prices were rising, hewas called, after the crash, the Pied Piper (or worse) of that episode. Yet, although he wasnot trained in any quantitative discipline, he seems to have been the only one to retainenough common sense and ability to comprehend the power of compound interest to realizesomething was wrong. Unfortunately for him, and a myriad of his followers, he drew the

wrong conclusion from his observation, and helped inflate the telecom bubble even further.The demonstration of the objectively irrational behavior presented in sections 4 through 8

leads naturally to Section 9, which briefly introduces a concept called information viscos-ity. This refers to the phenomenon that important information often does not spreadefficiently, and so does not get incorporated properly into market prices. This is contraryto the basic assumption of theories of efficient markets.

After the discussion of the irrationalities and defective information distribution in thetelecom bubble in sections 3 through 12, the rest of the paper considers briefly some relatedobservations, and their implications. Section 10 considers the general role of gullibilityin our society. Section 11 deals with bubbles and their detectability and controllability.Section 12 produces more examples of information viscosity. Section 13 relates gullibilityto various human traits, and presents the Madoff fraud as a positive indicator of the levelof trust in our society, the type of trust that is essential for the functioning and progressof modern economies. Section 14 contains a brief discussion of how increasing gullibilityis cultivated in our society. Finally, Section 15 has the conclusions, with some specificsuggestions on how a gullibility index might be developed.

2 Innumeracy

Gullibility is strongly correlated with innumeracy, the inability to reason with numbersand other mathematical concepts. Innumeracy is almost universal, and can be seen every


7/46


day. As just one example, the Wall Street Journal, one of the most prestigious businesspublications, managed recently to use $6 million in one place, and $6 billion for the same

quantity a couple of lines further [63]. Another example comes from a white paper fromIBM, one of the most eminent and most successful technology companies. It claims ( [28],p. 2) that by 2010, the worlds information base will be doubling in size every 11 hours,implying in a year there would be more bits of information than the number of elementaryparticles in the universe!

Innumeracy is especially dangerous in situations such as the telecom bubble, where thequantities under discussion are huge, with prefixes such as tera-, peta-, and exa-, and referto photons and electrons, objects that are not very tangible to most people. (That maybe part of the reason the Internet bubble fooled people so much more than the RailwayMania of the 1840s, which dealt with passenger transport.)

However, innumeracy appears to be only a part of the story. As we will see, many peoplewho otherwise exhibited substantial degrees of numeracy in their careers, to the point ofgetting PhDs from MIT and MBAs from Harvard, still fell for the most preposterouslyimpossible quantitative stories. And John Allen Paulos, the mathematician who wrote thefamous, illuminating, and for many readers frightening book Innumeracy [50], was also avictim of the telecom bubble [51].

Still, innumeracy offers one of the most promising methods to approach the studyof gullibility in a quantitative way. It appears (as will be shown later with numerousexamples from the Internet bubble) that as a mania advances, innumeracy grows, andgrows particularly dramatically among business and technology leaders, the people withPhDs and MBAs who are normally competent with basic arithmetic. Hence a quantitative

measure of innumeracy, which ought to be feasible, since it involves explicit numbers inreference to specific objects or services, could become a key part of the gullibility index.

3 Telecom bubble and Internet traffic doubling every 100days

Discussions of the Intenet bubble tend to concentrate on the dot-coms. They are usuallyheld up as examples of extreme irrationality. Yet they were by some measures the mostsober part of the Internet bubble, and were a great success. In retrospect, we can look back

and say that WebVan and eToys were silly wastes of investor funds. However, they did notconsume much funding. If we consider just the real investments of the dot-com boom, themoney spent on writing software, buying servers, as well as the legal, marketing, and otherexpenses, the total appears to be well under $20 billion dollars. Any single one of the (few)great success stories of that mania, namely Google, Yahoo!, eBay, and Amazon, createdmore value (from the perspective of todays market valuations) than the sum total of allthe misbegotten ventures. Thus from the standpoint of society as a whole, the dot-comfrenzy was a great success. A potentially different picture emerges when one looks at theprices paid for dot-com shares at their IPOs or later. But that represents just a transferof money from one class of investors to another, and not real economic activity. On the


8/46

8 Andrew Odlyzko

other hand, the real investments of the telecom bubble (and the associated more generalinformation and communication technologies bubble) were far larger.

The dot-coms were also hard to discount beforehand, at least with any solid arguments.Yes, eyeballs and mindshare were vague concepts, and it was not easy to see how toconvert them into revenues, but then it was hard to prove such conversion would not workeither. (It did work for Google, lets not forget.) On the other hand, what the Economistcalled embarrassing collective hallucinations of the telecom bubble were embarrassinglyeasy to show as destined to crash, largely because they were justified on the basis ofquantitative claims that were transparently false when describing the situation of that time,or absurd when describing the promised future. A passage from The Times (of London)from the time of the Railway Mania is singularly appropriate:

The salient points of absurdity in ... statistics were so numerous that we found it

impossible to notice more than a portion. It was tantalizing indeed to see so luxurianta crop, and feel that human hands could not grasp more than a handful.4

A fuller treatment of the patently false projections and claims will be presented in [49].In particular, that work will consider the financial aspects of that bubble, including theWorldCom accounting frauds. (The famous ones, the ones that resulted in Bernie Ebbersgoing to jail, were probably not the most damaging ones.) Right now, in the interestsof brevity, and since human hands cannot grasp more than a handful of the absurdnumbers and arguments that were used, I will concentrate on just one key issue, that ofInternet traffic growth. The preposterous nature of the claims made there is easiest todemonstrate, and had the most visible (and destructive) effect on actual investments. Italso demonstrates most clearly the irrational behavior of business and technology leaders,as well as of the press.

The key mantra of the telecom bubble was that of Internet traffic doubling every 100days. (Sometimes the 100 days was replaced by 3 months, or 4 months.) This mantra waswidely held, and was crucial in justifying construction of the new fiber networks. It wasalso a key inspiration to the the rest of the Internet bubble, including the dot-coms. As anexample, a story in the New York Times in March 2000, at the peak of the tech market,

just as NASDAQ shares were beginning their 80% decline, quoted Matthew Johnson, chieftrader of Nasdaq stocks for Lehman Brothers, on why shares were not overvalued [26]:

The hardest part for a lot of us on Wall Street who have taken traditional valuationmeasures, if you try to apply them today youd probably never buy a stock. Im abeliever in the new paradigm. Traffic on the Internet is doubling every 100 days;if you think about that, you begin to understand the magnitude of this technologyrevolution and you can understand investors willingness to take the risks theyretaking.

And indeed, had Internet traffic grown at that rate, not only would the telecom industryhave flourished, but so probably would have many of the dot-coms. The rapid growth oftraffic would surely have arisen from rapid adoption of new technologies and new onlinebusiness models, all progressing at the proverbial (but in fact horribly misleading) Internet


9/46


time. That would likely have produced bountiful revenues and profits for many of the newventures.

The claim that Internet traffic was doubling every 100 days was completely uncontro-versial, as far as the mainstream press was concerned, all the way to the end of 2000.During that year, the New York Times alone had at least five stories that cited it, in allcases as something that everyone knew was true5. The Nov. 27, 2000 issue ofFortune evenhad two stories, by two different reporters, that cited this myth. And there were manyauthoritative-sounding parties that helped propagate it. For example, in the conferencecalls with financial analysts to discuss the third quarter 2000 results, the heads of AT&T,Global Crossing, and Level 3 (and quite possibly others) claimed that Internet traffic wasgrowing that fast.

As seems typical, government officials were among the most gullible. Reed Hundt, theChairman of the FCC (Federal Communications Commission) from 1993 to 1997, claimedin his 2000 book [27] that [i]n 1999, data traffic was doubling every 90 days .... Anofficial March 2000 FCC document, submitted to Congress by the then-Chairman of theFCC, William Kennard, [30], claimed:

Internet traffic is doubling every 100 days. The FCCs hands-off policy towardsthe Internet has helped fuel this tremendous growth.

This language is eerily reminiscent (preminiscent ?) of that used by financial regulators adecade later, who hailed their hands-off policy as leading to the flowering of financialinnovation, in particular to the astronomical growth in volume of derivatives, which wassupposedly leading to a new era of prosperity and stability. The temptation to carry out a

deeper comparison of the underlying philosophies, non-actions, and eventual consequencesof these two episodes, separated by a decade, is almost irresistible. But in the interests ofbrevity, let us resist it for now.

We do have to observe, however, that the consequences of FCCs hands-off policy werecatastrophic for the economy. Note that a simple Internet traffic volume reporting require-ment, similar to the one that had been in force for decades for voice traffic, would havesufficed to disprove the myth and squash the huge spending on new fiber. It might alsopossibly have dampened the animal spirits behind the dot-com craze6.

The FCC document [30] is a reflection of FCCs gullibility and ignorance, but by itselfit does not seem to have had any significant influence. On the other hand, a document fromthe U.S. Department of Commerce, the 1998 white paper The Emerging Digital Economy

[61], was tremendously influential in stimulating the dot-com bubble. It was cited by in-numerable news stories and start-up business plans as supporting an extremely optimisticoutlook for new Internet-related ventures. This report in effect argued (combining thewords in the report to go slightly beyond what it explicitly said) that dramatic improve-ments in computing power and communication and information technology would createa long boom which [would] take the economy to new heights over the next quarter cen-tury. Among the 7 points cited on p. 2 of [ 61] as showing the growth of the Internet andelectronic commerce this past year, number 3 was the claim that [t]raffic on the Internethas been doubling every 100 days. This claim, as well as a similar, but more extensive oneon p. 8 of that report, cited a 1997 white paper from Inktomi. The Inktomi report in turn


10/46

10 Andrew Odlyzko

quoted Mike ODell, the Chief Scientist of UUNet, as saying that [t]he capacity crunch isreal and will continue for quite some time, and stated

UUNet estimates that network traffic is doubling every 100 days as graphics, au-dio, and video become more common. This rate means that bandwidth demandsare increasing at five times the rate of Moores Law. The ability to build networkcapacity cannot keep pace. ODell concludes, Demand will far outstrip supply forthe foreseeable future.

UUNet is engaged in a million-dollar-a-day capital expenditure plan to add thebandwidth needed to handle more users and increasing traffic loads. In three years,its network will be a 1,000 percent bigger than its current network.

The Inktomi white paper also provided a chart of Internet traffic growth, taken from theGilder Technology Report (which will be discussed at some length in Section 8).

There were several noteworthy features of The Emerging Digital Economy that appearnot to have been noted by the media nor by the numerous venture capitalists and en-trepreneurs who cited it. One is that the hard Internet traffic data from The Gilder Reportonly went through the end of 1996. When questioned on this issue, in an email pointingout that there was evidence of a dramatic slowdown in Internet traffic growth in 1997,one of the authors of the Department of Commerce report responded that the Internetis evolving so rapidly, that information quickly becomes out of date.7 That was certainlya paradox common to many Internet enthusiasts of the time. Even though they were al-most uniformly believers in the mantra of Internet time, the concept that everything waschanging many times faster than in the physical world, they would fixate on a statistic that

may have been true at some point, and keep repeating it as a settled fact for years, withoutchecking. (The doubling of traffic every 100 days was approximately true in 1995 and1996.) Another interesting feature of the Internet growth claims in The Emerging DigitalEconomy was that they were based on the Inktomi report, which demonstrated commoninnumeracy. The claim of a doubling every 100 days corresponds to growth of over 1,000%per year (more precisely, 1,155%, but that level of precision is irrelevant), and not thegrowth over three years of 1,000% cited in the Inktomi white paper. (Almost certainly, aswe will see in the next section, ODell had told Inktomi that the UUNet traffic was goingto grow 1000-fold in three years, meaning approximately 100,000%.)

Tales of fast growth of Internet traffic were the key element behind the telecom bubble.All the dark fiber that is lying around, still dark, was put down as a result of exaggerated

expectations of Internet traffic growth. Towards the end of the telecom bubble, in 2001and 2002, as the previous business plans were becoming less and less credible, telecomleaders often talked of how the fiber was just a way to deliver futuristic (but unnamed)services. Even some complaisant financial analysts had difficulty swallowing that story.Another explanation that was propagated was that the cost of lighting the fiber was manytimes the cost of laying the fiber, and so it was not wonder so much of the glass wasidle. This was true, but made nonsense of the financial plans. Since [t]he salient points ofabsurdity in the telecom bubble form so luxuriant a crop ... that human hands [cannot]grasp more than a handful, let us not explore financial aspects of the telecom bubble here,and concentrate on the basic traffic growth story. If we examine documents from the late


11/46


1990s, it is clear that this was by far the most important justification for the new telecomventures, in fact, almost the only one. See, for example, the presentation from 1998 by

Tom Soja [56], an important consultant to underwater cable companies. It was clear thatgrowth rates of traffic were the only thing that mattered for Global Crossing. Or consider360networks. At the end of 1999, it lured Gregory Maffei, the CFO of Microsoft, as its CEO,inducing him give up $64 million in Microsoft options (in return for a possible $886 millionin 360networks shares, were the new venture to succeed). Further [22,66], it had a technicaladvisory board consisting of the Whos Who of the New Economy, including Michael Dell;Nathan Myhrvold, chief technology officer for Microsoft; and Terence Matthews, founderof Newbridge Networks.8 The point is that it was to be a carriers carrier, providingbasic pipes to other service providers, expecting to capitalize on the amazing growth inInternet and data traffic. Thus in spite of the spin that it was attempted afterwards, thedriving force behind the greenfield fiber deployment was the perception that Internet trafficwas exploding, and would continue to do so, and it would be the provision of commoditybandwidth that was the road to riches.

Given so luxuriant a crop of absurdities, I do not deal with the many ways in whichfinancial projections for the telecom industry should have been seen as fatally flawed. Somemore detail on that topic will be presented in [49]. But for completeness, it might be worthmentioning that it would have taken something like a doubling of Internet traffic every 100days for the first half a dozen years of the third millennium for the greenfield fiber playersto make money. On the other hand, equipment providers and established telcos would likelyhave done extremely well even with 4x annual growth. What killed the entire industry wasInternet traffic growing about 2x annually, which only served to offset the technological

and architectural improvements in communications, and led to stagnant service revenues.

The truth was that while Internet traffic did double about once every 100 days in 1995and 1996, by 1997 the rate of growth had declined to doubling once a year, the rate that hadprevailed rather regularly in the early 1990s9. The FCC, the Department of Commerce, thepress, and the VCs and entrepreneurs and investors were all like Wile E. Coyote, happilyrunning effortlessly through the air, unaware they were in a freefall, very far from thehard ground beneath. Their oblivious attitude was enabled by a lot of hot air, emanatingprincipally from WorldCom and its UUNet branch.

The full story of the telecom bubble is complicated. Even the more detailed versionin [49] will only be able to present a simplified picture. But it should be said that in

addition to the nonsense from WorldCom/UUNet about growth rates, misinformation thatwas often amplified by others, there was much confusion and pressure to make decisionsquickly. Two participants in those events told me separately that there was no time tothink. Still, Internet traffic was known to be key to the health of the telcom sector, andpeople obviously did think about it, since they frequently cited the doubling every 100days story. Those who wanted to believe in it, could even occasionally find snippets of solidinformation from outside WorldCom/UUNet that supported it. For example, the respectedreports from the DellOro market research firm showed that the maximal total capacity ofall routers shipped in 1999 with ports of at least 2.5 Gbps increased by a factor of 7 overthe corresponding number of 1998. Similarly, the optical component markets sometimes


12/46

12 Andrew Odlyzko

provided examples of large jumps in demand10. In reality, most of those router ports tookseveral years to be populated, many of the optical component orders represented multiple

orders placed in anticipation of shortages, etc. And, of course, quite a bit of the demand wascoming from the new networks that were being built in anticipation of booming demand,networks that were often abandoned. As was revealed a few years later, actual Internettraffic growth rates varied, from 7080% per year at UUnet, to about 100% at Genuity,to about 300% at AT&T. Most estimates agree that growth in the year 2000 overall wasaround 100%, 2x per year.

In any event, the 10x annual growth rate claims should have aroused intense curiosity.Andy Grove in his famous book Only the Paranoid Survive wrote about the importanceof 10X change; whenever some important parameter changes by a factor of 10, one needsto fundamentally rethink business plans. The appeal of the doubling every 100 daystale of course was that it implied the world was now on a path of revolutionary upheaval.WorldCom talked of how their system architecture has to be redesigned not just boosted- every year. Every year UUNet redesigns the basic network architecture. (At the sametime, WorldCom was also telling investors that their big investments were over. With thenetwork buildout complete, they were about to to enjoy the bountiful profits their moneyhad earned. The discrepancy between this story and the one about annual redesign of thenetwork architecture was not noted at the time. I will not dwell on it here, since there is soluxuriant a crop of absurdities in the WorldCom story, and so I am leaving the financialsaside.) This was then extrapolated to the rest of the world by enthusiastic listeners. Butthe story was about 10X growth year after year. Those who retained the ability to thinkquantitatively, and had any sense for technology, were able to recognize such rates could not

be sustained for long. But there seemed to be few such people. It was shown explicitly in[45] that such growth rates would have implied that by the end of 2000, the average trafficper Internet user would have been over 1.5 million bits per second around the clock (whenmost people had at best 28 thousand bit per second modems they used for perhaps onehour per day, at a fraction of the capacity). Many readers, who had apparently managedto resist all the solid evidence of [8,9,55] that Internet traffic was only doubling once ayear, appeared then to finally grasp that their doubling every 100 days belief was adelusion. But why was it necessary to present that simple compound interest calculationexplicitly? It should have been obvious, and it had indeed been obvious all along to many.The most convincing explanation is that bubble participants lost the ability to comprehendcompound interest.

4 The preposterous ODell and Sidgmore lectures

The only seemingly credible direct source for the doubling every 100 days story was theUUNet division of WorldCom11. Unfortunately this source had great credibility, as it wasreputed to have the largest Internet backbone in the world, and at times claimed to becarrying half the worlds Internet traffic. The WorldCom acquisition of MCI and the lateraborted acquisition of Sprint were carefully scrutinized by competition authorities in theU.S. and Europe largely because of concerns the combinations would have monopoly powerin Internet services.


13/46


However seemingly-credible the source, the claims should have aroused suspicion earlyon. Not only were they the most extreme in the industry, but there were lots of indicators

that something was fishy about them. These indicators form so luxuriant a crop thathuman hands could not grasp more than a handful, so I will concentrate on the mostprominent and most extreme case, the public presentations by Mike ODell and JohnSidgmore of WorldCom/UUNet. But there were plenty of others12.

The most famous disseminator of the Internet doubling every 100 days tale was JohnSidgmore. At the end of the 1990s he was the Vice Chairman of MCI WorldCom, andafter Ebbers was forced to resign in 2002, was briefly CEO of that company. His obituaryin the New York Times noted that he had been one of the industrys most well-likedexecutives [13]. The obituary, printed after WorldComs bankruptcy and the revelation ofnumerous accounting irregularities at WorldCom, also noted that he was never chargedwith any misconduct. On the other hand, it is also known that his estate paid a multi-million dollar settlement in the class-action lawsuits that followed. As usual, the detailsare unknown, but it seems safe to assume that there was good evidence of negligence orworse on Sidgmores part that had been found. This may not have been connected at allto the tale of Internet doubling every 100 days. However destructive this tale was, andhowever false, and however fraudulently concocted, it is not clear that it broke any laws.Lying is not a crime, and this may have been just a case of the inactionable puffery thatour society tolerates (and, some would say, encourages). All the WorldCom insiders I havespoken to, ones who had been aware of the falsity of the doubling every 100 days taleand opposed to it, spoke very highly of Sidgmore. They thought that he had been simply

fed that line to push, and had been sufficiently far away from operations not to recognizeit as false.

Sidgmore seemed to be ubiquitous on the high-tech circuit, making presentations, oftenkeynotes, at various conferences, often run by investment banks or other organizationsfor investors and investment managers. Most of the news stories that did cite sources forthe astronomical growth myth mentioned Sidgmore. Unfortunately so far no recordings ofhis lectures have turned up (although there are copies of his paper from the Vortex 98conference floating around). On the other hand, for many years we had available a videorecording of a similar presentation by Mike ODell, Chief Scientist and Vice President atUUNet [42]. The context of this talk is almost as important as the talk itself, and will play

an important role in the discussion in Section 6. It was one of a dozen talks at a symposiumon Optical Internet: The next generation, held at Stanford University on May 16, 2000.(I.e., this was shortly after the dot-com collapse started, but while the telecom industry wasstill going strong. The flow of investment money was beginning to slow down, and priceswere drooping, but overall, spirits were still high.) It was associated with the inaugurationof the Stanford Networking Research Center, a cooperative venture of Stanford and (inthe words of the official web site) leading information technology corporations and SiliconValley industries. Attendance exceeded expectations, with many guests forced to watchvideocasts of the lectures in an overflow room. Participants apparently included numerousnetworking industry players, as well as venture capitalists, and of course students and


14/46

14 Andrew Odlyzko

faculty at Stanford. Thus this was about as sophisticated an audience as one could hopefor. And yet they fell for transparent nonsense of the ODell talk.

It has to be admitted that it was a wonderful talk, catnip for all the technology enthu-siasts in the audience. So it is a pity that the video of it has vanished from the Stanfordsite. (Hopefully a copy will be found someplace.) Still, we can get a taste, although aninadequate taste, from the transcript at http://www.dtc.umn.edu/odlyzko/isources/.The talk had some nice lines, such as the similarity of the Internet revolution to the disin-tegration of central planning, the difference between an eccentric and a madman is thatthe eccentric has a checkbook, and of course the final line, occupying the last slide, theline that had apparently been used by ODell and Sidgmore many times before, and wasoften repeated by others:

If you arent scared, you dont understand.

This was definitely (although inadvertently, since the intent was to convince the audienceto accelerate research, development, and deployment of networking technologies) true. Itappears nobody in the audience understood they were being fooled, the industry hadalready fallen off a cliff, and was about to crash. So they were enthused, not scared.

Why should the audience have been suspicious? Well, for one thing, ODell was pro-jecting growth by a factor of between 106 and 107 over the next 5 years, for annual growthby a factor of 16 to 25. And he was projecting petabit trunks, multiple ones. The au-dience should have recognized that this was totally impossible, at ODells time scale,with any conceivable technologies. But they swallowed the line, which was made tastierwith the usual techniques of warning people ahead of time it might seem impossible, etc.

(cf. [12,59]). And there were other obviously questionable aspects of the presentation. Forexample, ODell was describing how the UUNet buildout, to be completed by the end ofthat year, was going to use four times the worlds annual production of OC-192 lasers.Where were they going to come from (especially since few had been produced in previousyears)? Were they going to be imported from Mars?

However, predicting the future is hard, and miracles or almost-miracles do happen. Still,ODell also made some supposedly factual statements about the past that should have setalarm bells ringing. The second through fourth of his slides were the famous UUNet GlobalNetwork charts, which apparently also featured in all, or most, of the Sidgmore presen-tations. These three slides (available at http://www.dtc.umn.edu/odlyzko/isources/)purported to depict the UUNet network in mid-1997, mid-1998, and mid-1999. Tom Stluka,

who was involved in preparing these slides in early 1998, reports (private communication)that the network capacity shown for mid-1997 was correct, that for mid-1998 probablycame close to being installed on schedule, but the one depicted for mid-1999 was likelynot achieved until a year or more later. But they were all being presented in mid-2000 ashistorical facts! The audience had no way of knowing that, however. Still, by comparing thefigures for US Domestic Backbone capacity shown in the lower left of those slides, theycould have noticed that the jump from mid-1997 to mid-1998 was by a factor of 7.3, and inthe following year by a factor of 7, significantly less than the factor of 10 that ODell wasclaiming for each of the preceding 6 years. Even ignoring that, if one took the mid-1999figure of 268,794 OC-12 miles, and combined it with the ODell claim of 10 6 growth over


15/46


the previous 6 years, one discovered that his slides implied that UUNet had, in mid-1993,U.S. network of 0.28 OC-12 miles, which is equivalent to a single voice line across the

continent. That was impossible, as UUNet was already a large ISP back in 1993, with anextensive network of T1 (24 voice lines each) or faster lines across the country.

With a little digging for information one could also show that the ODell slides contra-dicted official UUNet filings. The S-1 form that UUNet filed with the Securities and Ex-change Commission on April 10, 1995, while it was still an independent company, statedthat its network then included a 45 Mbps ATM backbone13. Well, even a single trans-continental 45 Mbps link in mid-1995, combined with the stated capacity of 5,281 OC-12miles in mid-1997 shows that the growth over those two years could not have been higherthan a factor of 30, far short of the 100 claimed by ODell in 2000.

Still, as with the Sidgmore presentations, the audience for the ODell reacted neitherwith derision nor with indignation, but with applause. A recent passage about art fraudappears to describe the situation:

Forgers usually succeed not because they are so talented but, rather, because theyprovide, at a moment in time, exactly what others desperately want to see. Conjurersas much as copyists, they fulfill a wish or a fantasy. And so the inconsistencies crooked signatures, uncharacteristic brushstrokes are ignored or explained away.14

5 Capacity versus traffic

Section 3 discussed the role of the Internet traffic doubling every 100 days myth in the

telecom bubble. However, the reader may have noticed that the section above discussednetwork capacity, as measured in OC-12 miles, and that this is what is presented in theODell slides and in the transcript of his talk. This leads to another aspect of the story,one that provides more data on gullibility.

In the 1998 to 2001 period, ODell and Sidgmore appeared to always carefully talkof network capacity, not traffic. However, practically all press reports talked of traffic. Itappears that people heard capacity, but took it to mean traffic. A few, such as Jon Healeyof the Los Angeles Times, knew the difference, and were sensitive to what ODell andSidgmore said, but the general assumption was that the two were the same. (And, on thebackbones of the Internet, they are indeed not much different, certainly not when one haseven 2x annual growth rates.)

On a few occasions, WorldCom personnel appeared to talk of rapid growth in traffic. Afew examples are cited in [46]. Another is the story [24] from May 2000, the same monththat the ODell Stanford lecture was delivered, in which Joe Cook, WorldCom Inc.svice president of network systems engineering is quoted as saying that [n]etwork trafficvolume for [his] company is growing eightfold each year. However, those seemed to beexceptions, and it is hard to tell whether those WorldCom/UUNet staff forgot to toe theparty line, or whether the reporters translated claims about capacity into claims abouttraffic. As it turns out, UUNet traffic was about doubling each year during that period.This was implicit in a posting by ODell to a mailing list at the end of 2000, a posting tobe discussed below. It was also confirmed officially by the company in mid-2002, after its


16/46


17/46


6 Divergent growth projections, or 4 versus 10

Let us go back to the ODell presentation at Stanford, and consider the reactions of theaudience. They applauded, and there is no sign on the videotape of anyone raising anyquestions about the veracity or plausibility of ODells claims. That is another testamentto the power of the collective hallucination of the time, especially since the other events ofthe day provided plenty of grounds for raising serious questions.

Suppose it really is too much for people to remember figures from a slide flashed onthe screen, too hard to do compound interest calculation to show that those figures implysomething absurd about the past, and too hard to actually dig up SEC documents to showthat those figures contradict official filings. Even then there was plenty in the Stanfordmeeting at which ODell presented his talk to alert anyone with any sense, and not in thegrip of a massive delusion, that the telecom industry was in grave danger, as it did nothave any idea whatever about future demand for its services.

The ODell presentation was just one of three in the last of four sessions in the Stanfordsymposium that lasted all of May 16, 2000 (see http://www.dtc.umn.edu/odlyzko/isources/for more information). There were many statements about high growth rates throughoutthe meeting, some with specific figures. The remarkable thing is that those figures variedtremendously! Steve Alexander, Senior Vice President and CTO of Ciena, for example,casually mentioned 400% growth in two years, which is just 2.24x per year. Rao Arimilli,Vice President of Software Product Marketing at ONI Systems mentioned 35x growth infour years, which is 2.43x per year. Then, in the session just before ODells, Don Smith,the President of Optical Internet at Nortel, mentioned several times that his company had

done extensive studies, and expected 100 to 200-fold growth over the next four years, forannual growth rates of 3.16x to 3.76x19. Smith mentioned that his projections were higherthan those in earlier sessions. And then came ODell, with his 10-fold growth per year,and many explicit statements that this was far faster than any of the previous estimates.All the talks were applauded. Each session was followed by a question and answer period(about half an hour in each case, following on about an hour for the formal presentations).Some of the questions after Smiths session assumed explicitly his 100 to 200-fold growthin four years. And the questions for ODell all assumed the reality of his 10,000-fold growthprojection over the next four years. Nobody asked how real the projections were, nor didanyone ask about the wide discrepancy.

The inability to tell 4 from 10 was endemic. (Or at least the inability to recognize that

it meant a huge difference for the future of the telecom industry.) It did not reign juston the Stanford campus on the day of the ODell talk. The general press, as was shownbefore, was full of claims of Internet traffic doubling every 100 days. But if we consider

just January 2000, and look at the publications of IEEE, one of the premier professionalorganizations, we find different estimates. Larry Roberts, for example, one of the fathersof the Internet, claimed that month that Internet traffic was growing 4x per year in IEEEComputer [54]. An even more interesting case is that of IEEE Internet Computing. Thebulk of the January/February 2000 issue of this magazine was devoted to An internetmillennium mosaic, a set of predictions for the next decade of the Internet by a collectionof tech luminaries, including Bill Gates, Bill Joy, Leonard Kleinrock, and Eric Schmidt. Bob


18/46

18 Andrew Odlyzko

Metcalfe, the inventor of Ethernet and founder of 3Com, was then a technology columnistfor InfoWorld (and is now a venture capitalist). His piece declared that Internet traffic is

doubling every four months. Thus Metcalfe was claiming 8x annual growth, close to thedoubling every 100 days myth. On the other hand, Ross Callon, the Chief Architect forIronBridge Networks, stated in that same issue:

Bandwidth will continue to grow rapidly. The rate of growth will average a factorof 4 per year over five years (implying total growth will be roughly a factor of 1,000relative to 1999 bandwidth).

So here we had two well-connected technologists with greatly differing views on the mag-nitude of Internet growth. And yet there was no discussion about the difference, and whatit might mean.

Some more of the variety of claims about Internet traffic will be discussed in Section 9,and in much more detail in [49]. It appears that, in spite of the Internet traffic doublingevery 100 days mantra, meaning around 10x annual growth, that dominated the popularpress, much of the telecom supplier sector (meaning companies like IronBridge Networksand Nortel) was assuming about 4x annual growth. But there was no discussion of thedifference, nor any visible sign of it in the mainstream business press.

Simple compound interest calculations would have shown that at the growth ratesbeing discussed, even small differences could lead to catastrophic mistakes. At the Stanfordsymposium, ODell talked of the imperative need to grow the network 10x per year, andthe need to start deploying some systems three years ahead of demand. But at the 4xgrowth rate projected by Nortel at that meeting, in three years Nortel production capacity

would grow only 64x, while, according to ODell, UUNet network would grow 1,000x. Inother words, Nortel, which was at the time the dominant supplier of photonic systems tothe industry (after making a bold leap to a more advanced transmission system that theircompetitors had not thought would be needed), was at risk of being relegated to holding

just a 6% share of UUNet business in three years.Cognitive dissonance is the term (citing Wikipedia) for an uncomfortable feeling

caused by holding contradictory ideas simultaneously. Telecom bubble participants ap-peared to be able to transcend cognitive dissonance, by holding contradictory ideas simul-taneously without any uncomfortable feelings.

7 Underwater cables, and 2 versus 10

The collective hallucination of the telecom bubble was so powerful that it not only pre-vented observers from noticing that 4 is not 10, it even kept them from seeing that 2 isnot 10. There was a large and prominent sector of the telecom industry that based itsplans on an assumption of an approximate doubling of Internet traffic each year. Yet, withone singular exception noted later, no one seemed to notice, and it was the claims of 10xannual growth rates that dominated public discussion and planning.

The submarine cable industry, with enterprises like Global Crossing, Project Oxygen,FLAG, and Gemini, attracted much attention and many billions of dollars. Tom Soja was


19/46


one of the main consultants employed by this industry to estimate demand. First at KMI,and then at his own company, T Soja & Associates (TSA from now on), he prepared

the business cases for Global Crossing and several other underwater ventures. The slidedeck [56] from an April 1998 meeting shows the information on the data sources used tomake demand projections. The most conservative projection on slide 19 in [56], the 85%annual growth rate, was the key ingredient in the financial plans of Global Crossing andother carriers. It was also what Global Crossing told investors and the public that it wascounting on. For example, slide 10 of the presentation by Mool Singhi, the Director ofNetwork Planning for Global Crossing at the April 2001 conference [29] shows what itcalled Global Bandwidth Demand (Gbps) growing at an annual rate of 82% from 1999to 2004. An important point is that either implicitly or explicitly, these forecasts assumedthat bandwidth demand for underwater capacity would grow at about the same rate as forterrestrial capacity.

Financial projections are not the focus of this paper, but an obvious question arises.If Sojas traffic forecast was approximately correct, with traffic about doubling each year,how come Global Crossing went bankrupt? There are two interrelated reasons. One isthat Soja did not appreciate how quickly technology would improve, and he did not getthe financial dynamics right. And part of the unanticipated financial dynamics was therise of many competing projects. Soja (personal communication) thinks that many of thebankers (and there were hundreds at the numerous presentations he made) looked at hisprojections, saw all the other, far more optimistic growth scenarios in his charts (with themost optimistic one in slide 19 in [56] showing 300% CAGR), and assumed the true valuewould be someplace in between, and would provide plenty of room for other carriers tomake profits. It is quite plausible that if no other trans-Atlantic cable had been built bynew players, ones not affiliated with the established telcos, then the Global Crossing AC-1cable would have been very profitable. (But then the traffic would likely have fallen shortof Sojas projections, as prices would have been higher than they turned out. Fortunatelywe dont have to worry about such details here.)

TSA was not the only consulting firm making conservative, and, in retrospect, very rea-sonable, traffic projections. Arthur D Little (ADL) was also in that camp. Arthur Solomon,formerly a Vice President and Director of that company, and head of its telecommunicationsindustry consulting group, provided (private communication) a copy of the trans-Atlantictraffic forecast that his group had made for a client in 1999. It assumed growth rates declin-

ing from 76% in 1998 to 21% in 2009. The mix of applications it envisaged can be seen, inretrospect, to have been incorrect. (And so was the pricing forecast.) But the total trafficvolumes turned out to be extremely accurate. For example, for 2008, the forecast envis-aged average traffic of 859.2 Gbps. The estimate made by Telegeography in 2008 was ofaverage traffic in the spring of that year of 861 Gbps! Now this amazing coincidence shouldnot be taken too seriously. The ADL projection was for total traffic, including voice, andnon-Internet data networks, while Telegeography measured only Internet traffic, and forApril of 2008. (At the growth rates of Internet traffic, there is substantial variation evenfrom month to month.) Still, by any standard, actual volumes were similar. Further, theprojections made by the planners of the unnamed cable involved in that study were only


20/46

20 Andrew Odlyzko

about 20% higher than those made by ADL. (Solomon reports that in some other cases,cable promoters had far higher traffic expectations than those of ADL.)

The main point of these examples is that ADL, TSA, and many undersea cable promot-ers had expectations for Internet traffic growth that in retrospect were very reasonable. Inother words,

none of them drank the ODell/Sidgmore/WorldCom/UUNet Kool-Aid

of Internet traffic doubling every 100 days. They were assuming approximately 2x annualgrowth, in capacity and traffic, on both terrestrial and subsea links.

Enthusiastic financial analysts, with the now-infamous Jack Grubman in the lead, hadno difficulty acclaiming simultaneously the glowing prospects of WorldCom, with its 10xannual growth projections, and Global Crossing, with its 2x estimates. The discrepancy

does not appear to have been noticed in any publications I have seen, with the singularexception of the Gilder Technology Report.

8 George Gilders pioneering discovery that 2 is not 10

It is noteworthy that it was Gilder who noticed that 2x and 10x annual growth figures dontjibe. Gilder had no PhD from MIT, nor an MBA from Harvard, nor did he have a tenuredpost at a major research university. On the other hand, he was the greatest cheerleaderfor the Internet bubble in all its manifestations. Yet he seems to have been the only oneto have the wits to realize the various Internet growth projections were inconsistent, and

that this had serious consequences. Unfortunately, he drew the wrong conclusions from thisobservation, and this was instrumental in destroying his career, as well as in ruining manyinvestors.

Gilders role in the Internet bubble deserves a book-length study. As it is, we have tomake do with the informative but limited article The madness of King George by GaryRivlin in Wired [53]. Gilder was tremendously influential in stimulating investor interest intechnology stocks, and in affecting the thinking of business and technology readers. Evenmany of those who thought him close to the lunatic fringe paid attention to his speechesand writings, simply because so many others were doing so. At the peak, his flagship GilderTechnology Report, GTR from now on, had about 70,000 subscribers and brought in about$20 million a year in revenues [53]. The famous Gilder effect had share prices of obscure

tech startups zoom into the stratosphere after they were added to his list of promisinginvestments.

One thing that everybody agrees on is that Gilder is one of the most honest peoplearound. (That he kept some of his doubts about Internet bubble high fliers to himself, asrelated in [53], is a reflection of the skewed, some would say perverted, attitudes that seemto affect almost everybody at the peak of any mania.) Seldom in doubt, very often wrong,he is unquestionably sincere. He personally lost a lot materially in the Internet bubblecrash (aside from the damage to his reputation). Further, he is not a fool, as over the yearshe has had a number of brilliantly prophetic technological insights, some of which still havenot been absorbed by the information and communication industries, nor by the research


21/46


community. Thus he is another example of the difficulty of placing important bubble figureson the spectrum from rogue to fool.

Gilders fame and career in the 1980s and early 1990s was focused on computing, whathe called Microcosm. Then he realized that communications was the new frontier, andconcentrated on what he named the Telecosm. The monthly GTR, which started appearingin July 1996, devoted great attention to Internet traffic, which it called the real index ofInternet expansion (GTR, Sept. 1997, p. 2). The very first issue had a big chart of Internettraffic measures on its front page, and later issues continued this attention. (Much of thework in tracking such measures was apparently done by Ken Ehrhart, the Director ofResearch at the Gilder Technology Group, who had his name jointly with Gilder on someof the published pieces. However, for simplicity, I will write as if GTR was just Gilderswork.) GTR was the first at least semi-public work to systematically collect Internet trafficstatistics after the phase-out of the U.S. government-funded NSF backbone in early 1995,far ahead of [8,55]. The hard data in GTR unfortunately was based just on the trafficthrough the U.S. public Internet exchanges, the MAEs and NAPs. The statistics in GTR,if read with even a mildly open mind, showed a dramatic slowdown in traffic growth in1997. However, Gilder managed to gloss that over, assisted by the widely acknowledged butnot quantified fact that growth was faster at private exchanges for which traffic statisticswere not available. One could write a long article just about GTR reports and opinionson Internet traffic. But when one reads the back issues of GTR, what comes across isGilders struggle to understand what was happening, trying to reconcile fragmentary andconflicting snippets of information with each other and with his hopes and expectationsfor revolutionary change. Thus we find him writing of

MCI reports of traffic on their network growing by 6% per month (GTR, Sept. 1997,p. 5, which corresponds to traffic only doubling in a year)

The Law of the Telecosm, that each year will bring a 34 fold increase in bandwidth(GTR, Jan. 1998, p. 4),

Alan Taffel of UUNet being up against the kilofold wall, a thousandfold rise in networktraffic every three years (GTR, Dec. 1997, p. 1, which corresponds to traffic growing10x each year, the doubling every 100 days myth)

There is quite a bit of innumeracy mixed in, numbers that are clearly not consistentbeing tossed together20. There seems to be some initial skepticism about the World-Com/UUNet growth claims. Thus we find Gilder writing of how Alan Taffel of UUNet

makes a truly stunning claim about Internet traffic, confirming my most extreme projec-tions. At WorldCom-UUnet, so he says, traffic is increasing at a rate of some ten times peryear, (GTR, March 1998, p. 3). But as time goes on, the temptation to accept the World-Com/UUNet story proves irresistible, and we find Gilder writing without any hesitation ofdoubling of Internet traffic every four months.21

And then Gilder discovered that 2 is not 10, and that this mattered. Up until the endof 1998, Gilder did not pay much attention to submarine cables, and his darlings werethe terrestrial carriers, in particular WorldCom and Qwest. But then he looked at GlobalCrossing, and had a revelation. Gilder discovered that its business plans were based on amere annual doubling of traffic each year, and that undersea capacity had grown at lower


22/46

22 Andrew Odlyzko

rates over the preceding decade than terrestrial one. (He confused potential capacity offiber when lit with state-of-the-art equipment with actual used capacity, but thats another

story.) Hence he espied fantastic profit opportunities for Global Crossing. The Nov. 1998issue of GTR was entitled Cosmic Crossing: 1998s best opportunity, and argued that

undersea capacity increased some 42 fold since 1990 and will rise another 82 timesover the next three years. Thats a total of 3,444 times. That means that between1990 and 2001, terrestrial capacity will have increased by a thousand times morethan undersea capacity. ...

Assuming that global Internet traffic will prove to be growing at less thanmillionfold every six years projected by UUNet, the expansion will still be huge. ...

... undersea traffic will grow several times faster than terrestrial traffic. Takemy word for it. Over the next five years, the submarine portions of the Internet

will prove to be an agonizing choke point. Thus Global Crossing has a truly cosmicposition as the supplier of the missing element that completes the global system.

Gilders analysis showed considerable sophistication, far greater than that of most contem-porary observers. He realized that Internet traffic was very likely to become more local withtime, following the pattern of other communication technologies. (Not understanding theimportance of locality has hobbled many people over the ages, and was a key element in thefinancial disaster of the British Railway Mania of the 1840s [47].) Still, Gilder had enoughof a sense for the power of compound interest to realize that doubling each year (as inGlobal Crossing business plans) would lead to growth by a factor of 8 in three years, whiledoubling every 100 days would produce growth of at least 1,000 times over that period.

Hence if the WorldCom/UUNet doubling every 100 days story were true, even the mostextreme imaginable localization of traffic would still produce demand for subsea capacityfar in excess of the 8-fold growth anticipated by Global Crossing22. Therefore he became araging bull on the shares of this company. Gilder stayed a bull on Global Crossing until itsbankruptcy in January 2002, a bankruptcy that took Gilders reputation and his fortune(as we as those of many of his followers) down the tubes.

What Gilder appeared never to have entertained seriously was the idea that the dou-bling every 100 days story was an outlandish fantasy. Thus this presents a wonderful,if pitiful, example of extreme gullibility. Moreover, Gilder demonstrates the durability ofdelusions. As late as 2005, he could not bring himself to say that the WorldCom/UUNettale was a fraud, and was writing about an alleged Worldcom lie that Internet traffic was

doubling every 100 days, in the familiar pattern of people reluctant to admit they havebeen duped23.

9 Information viscosity

A foundational assumption of the efficient markets hypothesis is that asset prices in liquidmarkets incorporate all relevant information very quickly. (There are several versions of thehypothesis, let us not get involved in technical details.) Recent events show this is clearlynot true, as will be discussed later in Section 12. However, that was also not true duringthe Railway Mania of the 1840s [47], and it was not true during the telecom bubble. The


23/46


simplest example from this recent episode is the one discussed in the preceding sections,in which wildly differing growth rate assumptions were held by different segments of the

industry without the public being aware of this.As usual in bubbles, in the telecom mania there were from the beginning various skep-

tics and doom-sayers. The GTR of Sept. 1999 reported that its online Forum writhes andwriggles with bandwidth glut anxiety. Telecom consultant and financial analyst reportsestimated the then-current and future Internet growth rates at anywhere from the unreal-istically low 30% per year up to the WorldCom/UUNet fables 1,000% per year. (A moredetailed description of the various estimates will be presented in [49].)

There were numerous people who knew the truth. Tom Stluka and a group of otherWorldCom employees, embarrassed by the myth and concerned about the damage it coulddo to WorldCom, had hard direct knowledge from company data that the myth was false.Those outside could not be certain, but often had very convincing evidence that Internettraffic was not growing astronomically. For example, Scott Marcus, who was CTO of GTEInternetworking (which became Genuity), one of the Tier-1 ISPs of the late 1990s, reports(private communication) that he and his colleagues were quite sure, from marketing dataand statistics on traffic exchanged with other carriers, that both UUNet and entire Internettraffic was not growing much faster than a doubling once a year. However, they kept quiet,since

our firm could not benefit by publicizing the emperors lack of clothes. The mythwas well entrenched by then. We felt that the public would assume that our trafficwas growing more slowly than the norm, and that this perception would hurt us inthe marketplace.

So the engineers at Genuity had a good sense of the underlying reality. Oursenior management knew as well, but occasionally the marketing types seemed toforget. Or perhaps they did not completely believe their engineers, in the face ofthe steady droning from the trade press about traffic for our competitors that wassupposedly doubling every hundred days.

At AT&T, top management believed the 10x annual growth myth. AT&T salespeople toldcustomers that Internet traffic was growing that fast, and, when pressed, revealed thatAT&Ts own traffic was growing about 4x per year (which was true, as AT&T was goingall out for market share). But many AT&T engineers knew that the 10x fable was false,and that general Internet traffic was not growing much faster than 2x per year, even while

others continued to hold onto the doubling every 100 days tale.Many other people in the telecom industry knew the doubling every 100 days myth

was false, with various degrees of certitude. For example, much of the telecom suppliersector seemed to be planning on 4x growth, in some cases on less, which implicitly meantthey did not accept the ODell and Sidgmore fables. (Note that for suppliers, traffic is notdirectly relevant, it is capacity they get paid for. Thus it is high rates of growth of capacity,what ODell and Sidgmore talked about, that mattered.) There will be more examples anddiscussion of this topic in [49], but for the moment let us just note that, just as in theMadoff Ponzi scheme, the negative information was quite widely dispersed, was not held asa great secret, yet it did not affect press coverage, nor market valuations nor investments.


24/46

24 Andrew Odlyzko

There were many snippets of reasonably hard data to tell investors and the public thatthe rapid growth of Internet traffic in 199596, at about the proverbial doubling every

100 days rate, had come to an end in 1997. For example, the data released occasionally byMCI, one of the largest Internet backbone providers, showed this, as can be seen in issues ofGTR or in slide 14 of [56] (although both these sources put a more positive interpretationon the data than it warranted in retrospect). And then there was an explicit email, postedon a public web site, from Vint Cert, one of the fathers of the Internet and the MCIInternet guru, declaring at the end of 1997 that

traffic in our backbone is now at 100% per year compounded annually. Others reporteven higher values. Last year traffic increased by 500%.24

Then came the papers [8,55] with extensive data and documentation as to where it camefrom25. Both were widely read. The circulation of the Business Communications Reviewin 1999 was around 13,000, with about half enterprise users, largely telecom/IT managers,with some in upper management through the CIO level, and a substantial fraction amongtelecom service providers, and some on Wall Street26. The paper [8] was released for out-side distribution by AT&T in early July 1998, and was immediately placed on a publiclyaccessible web server, and announced on various mailing lists, resulting in around 2,000downloads in 1998. It was also the most frequently accessed paper on First Monday in1999, with over 28,000 downloads27. However, these two papers [8,55] seemed to have nonoticeable effect. The Sevcik work did attract a brief note in Upside magazine [15], butthis note did not convey the main import of [55]. Otherwise the world just went on itsmerry way, until around the middle of 2000, when a sense of alarm started developing, and

reporters and financial analysts started looking around for solid information about Internettraffic.

An interesting perspective on information diffusion is provided by Hui Pan and PaulPolishuk of Information Gatekeepers, a telecom consulting firm. They are certainly wellconnected in the industry. However, what prompted them (private communication), aroundthe middle of 2000, to organize the April 2001 conference on fiber glut [29] was not hardknowledge that there was a big mismatch between supply and demand. Instead, they justheard various expressions of unease and concern about the large number of players jumpinginto the industry. Another, similar, perspective, is provided by the famous mid-2000 memoof Leo Hindery, the CEO of Global Crossing, that his company and its competitors werelike the resplendently colored salmon going up river to spawn, at the end of our journey

our niche too is going to die rather than live and prosper. 28 He also apparently did nothave hard facts about the disaster that was coming, and was only deducing this fromthe behavior of competing firms. This lack of awareness of publicly and easily availableinformation is at wide variance with the assumption of the efficient market theory thatinformation diffuses quickly.

The telecom bubble also shows the limited reliance one can place on the press to dig upand distribute information contrary to the accepted wisdom. Reporters did not notice themany discrepancies and implausibilities in the stories about Internet traffic, and for a longtime ignored serious published data. Many reporters, even when prompted to investigate,had no interest. When one got intrigued enough to write a story in May 2001 about the


25/46


falsity of the Internet traffic doubling every 100 days fable, she emailed that she hadsubmitted a piece to [her] editors, but they havent had the space to run it. (This was a

year before the WorldCom bankruptcy, and an inquisitive look at the traffic myth mighthave led to the accounting frauds that destroyed the company, and so might have broughtthat house of cards down a year earlier.)

The general picture from the telecom bubble is similar to the one from the Madofffraud, and also from the British Railway Mania of the 1840s. The herd instinct dominates,and there is great reluctance to question the accepted wisdom and upset the status quo.This is contrary to the efficient markets theory, but very similar to what has been observedin politics. There, too, splinter groups get essentially no attention, until either they growbig enough, or some prominent person or institution takes notice of them, or else somegiant disaster takes place. That politics is similar to business should perhaps not be toosurprising. Both depend on tales, beautiful illusions, as well as credulous simplicity.

Information viscosity will be considered again in Section 12. First we consider somemore general issues of how our society and economy function.

10 Ubiquitous and deep gullibility

Many people express incredulity at the repeated failures of regulators to shut down theMadoff and R. Allen Stanford scams, in spite of the extensive evidence that was accumu-lating to show their enterprises were fraudulent. However, it is not absolutely necessary toinvoke malfeasance or perverted ideology to explain this phenomenon. Credulous simplicityis pervasive in our society (and, as will be argued in Section 14, may be vital to its func-

tioning). Just consider the Nigerian 419 scams, with emails that promise you 30% of $14.2million dollars, say, if you help the daughter of Jonas Savimbi move the entire amount toyour country. The name of this scam comes from the section of the Nigerian penal codethat covers it. The scam itself is old, predating the Internet. So one would think that bynow, all the people naive enough to fall for it would have been fleeced, and the scammerswould have nothing to gain. Instead, the scam flourishes, and in 2006, the U.S. SecretService estimate[d] that 419 swindlers gross[ed] hundreds of millions of dollars a year,[70]. Contrary to popular impressions, educated and supposedly sophisticated people areoften among the most gullible. We see that in the numerous fads in academia, and we seethat in the account in [70] of a Massachusetts psychotherapist who fell for the Nigerian419 scams.

Other examples of extreme gullibility among trained professionals abound. The JeromeKerviel affair, which cost the Societe Generale bank about $7 billion dollars, and brought itto the brink of ruin, revealed extensive management failures and weaknesses in the banksrisk-control systems, which had allowed at least 74 alerts about Mr. Kerviels activitiesto go unheeded over the course of 18 months, [6]. The Madoff affair provides multipleadditional examples. Victims were lured by a variety of methods, including the appeal ofbeing in an exclusive club, with threats of expulsion if they became obstreperously inquis-itive, or, in cases of some large and sophisticated investors, apparently by the implicationsthat the profits were coming from illicit trading [41]. Thus the Madoff scheme built onseveral well-known principles that scams are built on, cf. [12,59]. (It will be argued in


26/46

26 Andrew Odlyzko

Section 14 that these principles, as incorporated into legitimate practices, may be key toeconomic progress). But perhaps the most interesting instances of the Madoff fraud were

the numerous trustees of charitable and educational institutions who entrusted to him theendowments they were in charge of. These were supposedly sophisticated individuals,who clearly had the best interests of their institutions at heart. Yet they failed to do evenbasic due diligence.

And then there were regulators, private and government. Between 2003 and 2005, [thepredecessor of FINRA, the financial industrys independent watchdog] received credibleinformation from at least five different sources claiming that the Stanford CDs were apotential fraud, ([14], p. 2). Also, [n]obody was more surprised that the Securities andExchange Commission did not discover Bernard L. Madoffs enormous Ponzi scheme yearsago than Mr. Madoff himself, [25].

At the pinnacle of the regulatory establishment we find three professional economists,Alan Greenspan, Ben Bernanke, and Larry Summers. They were among the most gullibleof a very gullible profession. Alan Greenspan has been mentioned already. He is actuallycapable of learning. The Internet crash of a decade ago convinced him that accountantsneeded to be regulated [19]. And the 2008 crash famously convinced him that bankers couldnot be trusted either. Shareholders and taxpayers might complain about the length andespecially cost of these lessons, but they do show his ability to learn. The more interestingquestion is how come somebody so naive could stay at the helm of the worlds mostinfluential central bank for two decades without anyone in authority taking note? Is itperhaps an indication of the systems bias towards credulous simplicity?

Greenspan [21] has tried to defend his inaction and lack of curiosity about the dangersof the real estate and finance bubble by arguing that there are always dangers and thereare always doomsayers. However, what we had during the last decade was a combinationof developments, each of which should have raised warning flags. Total debt (as fractionof GDP) reached record levels, higher than in 1929. Housing prices (relative to householdincomes) reached records. Corporate profits soared to record levels, even while long terminterest rates were very low. Balance of payments deficits were higher than ever before. Fi-nancial industry profits as fraction of total profits reached unprecedented levels. There wasan explosion of derivative trading, without any commensurate effect on societys prosperity.And so on. Yet no serious investigation was undertaken by the Fed. (And, in a noteworthy

development, there was little curiosity about these developments among economists.)Greenspans dogmatic conviction that there was no need to be concerned about bubbles

was fortified by some much-cited academic papers of Ben Bernanke. A decade ago, beforehe became the head of the Federal Reserve, Bernanke published (together with MarkGertler) a paper which claimed, for example, that [a]dvocates of bubbles would probablybe forced to admit that it is difficult or impossible to identify any particular episodeconclusively as a bubble, even after the fact [3]29. Ben Bernankes contributions to thebenign neglect of bubbles was not limited to his academic work with Gertler. His manyofficial pronouncements, as head of the Fed, in the runup to the crash of 2008, that thefinancial system was in good shape do not need retelling.


27/46


Larry Summers contributed to the recent events in multiple ways. At the the Treasuryin the Clinton administration, he apparently played a key role in preventing regulation of

derivatives. Afterwards, at Harvard, he seems to have pushed that institutions endowmentinto some disastrous investments. And, when Raghuram Rajan, one of the few economiststo take a serious look at dangers developing the financial system, presented his findings,Summers publicly found the basic, slightly lead-eyed premise of [Mr. Rajans] paper tobe misguided, [32].

The aim here is less to point out mistakes made by these three economic policy makers,and more to observe that their views were well known when they were placed in positionsof power. Moreover, two of them, Bernanke and Summers, are now in charge of much ofU.S. economic policy, and of carrying out reforms. Can they be expected to do that well?Are they even truly expected to do so? After all, the political process that had placed them

in positions of power years ago is still in operation. Greenspan adamantly refuses to admitany serious fundamental mistakes [21]. Bernanke and Summers have not produced similardocuments, outlining their views on the recent crisis, but then they have not acknowledgedany serious mistakes either, and have not apologized for their actions. Further, it appearsthat the papers [3,4] have not been retracted (even though Bernankes actions as Chairmanof the Federal Reserve effectively repudiate the conclusions of those papers).

The studious efforts of the economics profession to avoid any serious investigation of theInternet bubble are also consistent with the view that our society is wedded to cultivatinggullibility. One finds a few hints in the literature on that mani

Bubbles, gullibility, and other challenges for economics, psychology, sociology, and information...

Documents

Transcript of Bubbles, gullibility, and other challenges for economics, psychology, sociology, and information...